Thunderbolt Card for 2010 Mac Pro

[Mr. Retrofire]

Originally Posted by odinsride

Has anyone heard any news on this? When they're coming out, how much they will cost, etc?

Dying to get my hands on the 27" Thunderbolt display!

First signs:
http://www.hardocp.com/news/2012/06/...rd_at_computex

;-)

----------

Originally Posted by MacVidCards

Isn't it funny/sad that this thread is a year and a half old, and we still don't have any idea if/when we willl ever see this?

Gigabyte motherboards have 2 Thunderbolt ports:
http://www.gigabyte.com/microsite/30...underbolt.html

iLOL

[goMac]

Originally Posted by Mr. Retrofire

First signs:
http://www.hardocp.com/news/2012/06/...rd_at_computex

That requires special support on the motherboard, but does at least show what Apple could do on the Mac Pro. Apple might even have some clout to put put the header connection onto the GPU instead.

Edit: Wait, blah, a GPU wouldn't need a header at all.

[ActionableMango]

I honestly don't think 2012 and older Mac Pros will ever get Thunderbolt.

[deconstruct60]

Originally Posted by BitterCreek

Please quote a source, wikpedia is not accurate for any form of information.

Here is a block diagram from a Intel motherboard with Thunderbolt.


[ Anadtech ran an article covering this another board. http://www.anandtech.com/show/5884/t...p8z77v-premium ]

The Thunderbolt controller is hooked to the PCH. This notion of direct coupling is goofy since Sandy Bridge/Ivy Bridge CPUs don't even have graphics direct output. There is not direct coupling to the CPU at all.

Likewise the Tomshardware article pointed to earlier in the post you are trying to knock down also has a diagram which probably more so illustrates the delusion flaw in your claims about direct CPU coupling.


[ Contextual page. http://www.tomshardware.com/reviews/...gd80,3205.html containing the borrowed from Intel graphics. ]

The first option above is all that is needed to make it work in a Mac Pro. The dGPX does not need to be implemented on a detachable PCI-e card. The iMac and MBP 15" models do did it in the 2011 & 2012 without one.

There is also no requirement that the TB controller pick up a display port output from the PCH if there isn't one or that there even be two Display port inputs (some TB controllers don't even have more than 1 input.). The PCH of a new Mac Pro could either use x4 PCI-e from CPU or x4 from its PCH. There is no requirement only to use one or the other.

There are limitations in placing the Thunderbolt connector. It pragmatically needs to be only a couple of inches from the connector ( the very high bit rates relative to even PCI-e make it a problem to place elsewhere).


Pragmatically though there is another limitation. Namely the Thunderbolt controller needs some power and firmware connectivity


[ Again the Thunderbolt 103 section in same article http://www.tomshardware.com/reviews/...80,3205-4.html ]

The "native host interface" (NHI ) + Thunderbolt Target mode (on Macs ) + the power distribution + "host mode configuration at boot"requirement means these 'custom' host interface on ASUS (and other ) designs for TB cards are more than just a simple DisplayPort connector from the motherboard's PCH DP output.

The block schematics are more about movement of data than of boot/firmare/power specifics. I think that is the tipping point against "plug in cards" for legacy oriented design around PCI-e cards. Not the PCI-e and DisplayPort data traces.



Impossible means can't be done at all; even as a kludge. That is fundamentally different from improbable ( extremely not likely). A kludge that hooks up legacy PCs (including previous Mac Pros ) could be done. As I pointed out before they'd have to kludge around the missing firmware/power/boot issues. It just isn't very likely. It is far more cleaner and easier to pass TB certification to run embedded GPU output along with the other connections needed from the motherboard.

[deconstruct60]

Originally Posted by goMac

That requires special support on the motherboard, but does at least show what Apple could do on the Mac Pro. Apple might even have some clout to put put the header connection onto the GPU instead.

Externally mounted loop back cables? Apple 'could' do that. They extremely likely will not. It is not a particularly simple or elegant design. Again the is a gap between possible/impossible and probable/improbable.

Apple could also so some custom GPU card to perhaps clean up the cables. However, they'd be apple proprietary GPU cards. If you thought the"Apple tax" on regular cards with Apple EFI support was high, that would likely be higher.

The Mac Pro desperately needs to get away from an overly narrow set of cards to survive long term. Custom workarounds will kill off the product long term.

[goMac]

Originally Posted by deconstruct60

Externally mounted loop back cables? Apple 'could' do that. They extremely likely will not. It is not a particularly simple or elegant design. Again the is a gap between possible/impossible and probable/improbable.

Apple could also so some custom GPU card to perhaps clean up the cables. However, they'd be apple proprietary GPU cards. If you thought the"Apple tax" on regular cards with Apple EFI support was high, that would likely be higher.

The Mac Pro desperately needs to get away from an overly narrow set of cards to survive long term. Custom workarounds will kill off the product long term.

I was thinking internal loopback cables. They have the influence on the GPU design to make it happen. Put a few internal DPs on the card and you're in business. But it would be kind of the opposite of what that ASUS board did. Loopback from the GPU to the Thunderbolt controller on the board.

But yeah, as you mentioned, it would mean more proprietary cards. But if things went well, maybe it would become a wider standard if PCs adopted the same sort of Thunderbolt implementation as well.

[deconstruct60]

Originally Posted by goMac

But yeah, as you mentioned, it would mean more proprietary cards. But if things went well, maybe it would become a wider standard if PCs adopted the same sort of Thunderbolt implementation as well.

Sounds like what was said when Apple created ADC ( http://en.wikipedia.org/wiki/Apple_Display_Connector ). It went no where as an industry standard. Apple adopting something that swims "upstream" against Thunderbolt and industry trends is extremely likely not to be adopted by anyone else.

Whether Thunderbolt is broadly adopted by the general PC industry depends upon how it is utilized with new systems and peripherals; not legacy designs.

The vast majority of new desktops have GPUs embedded in them now. Intel's and AMD's core desktop offerings references motherboard designs have GPUs embedded in them. It involves creating no new additional industry standard to wire up a Thunderbolt controller on them. For example, Apple's recent two iMacs solved the issue of connecting a dGPX to the controller with no new standards. So it is not even a solution that is foreign to Apple.

The Rube Goldberg complexity is only invoked when trying to push Thunderbolt into the past as opposed to the future.

[dyn]

Thunderbolt can not be compared to ADC at all. ADC is Apple's own technology but Thunderbolt is Intel's technology. Intel doesn't benefit from Thunderbolt being Apple-only, it needs everybody (Apple, the pc brands) to adopt it just like their USB3 standard, CPUs and many other products/technology. Apple aims for itself, Intel aims for anybody.

[SuperCyborg]

Originally Posted by goMac

What? Of course it is. It's just not a very complicated one.

Just because it looks like a CPU doesn't mean it is a CPU.
Would you call a north/southbridge a CPU?

And you still haven't said why it requires a CPU. It requires DMA

Guess where the memory controller is on every modern computer ? Right, on the CPU!

Intel originally demoed Thunderbolt from a PCI Express card, even.

On a logicboard specifically designed to demo it and it was not a PCIe card.
Ever heard of cost management?
Its cheaper to R&D on a removable card than a logicboard. That way, only the card has to be remade vs the entire logicboard.

Logical, huh?

And video output has nothing to do with the CPU.

TB uses the GPU built into the CPU.

On the iMacs, that's linked to the GPU, not the CPU.

Please provide proof to support your opinion.

----------

[/COLOR]

Originally Posted by ScottishCaptain

Then someone like Sonnet *could* make a card that either:

They won't because Apple and Intel hold the patents and aren't sharing/licensing the technology.

There is no physical impossibility.

Yes there is. Its called hardware limitations.

A third party company simply hasn't bothered to do it

Oh, yet, third parties have "bothered" to make Lightning cables that cost as much as the genuine Apple-brand?

----------

Originally Posted by Mr. Retrofire

First signs:
http://www.hardocp.com/news/2012/06/...rd_at_computex

Not a sign.

"With this in mind ASUS is proud to unveil its exclusive Thunderbolt upgrade solution for its line of Z77 and H77 motherboards."

Its not an expansion card that goes in a PCI-E slot, its an upgrade card made specifically for two brands of logicboards.

Gigabyte motherboards have 2 Thunderbolt ports

That is a built-in feature, not an expansion card.

[goMac]

Originally Posted by SuperCyborg

Just because it looks like a CPU doesn't mean it is a CPU.
Would you call a north/southbridge a CPU?

Image

Sure would!

It's not a general purpose CPU, but it's a CPU. That's why the distinction exists.

A North Bridge has a clock rate, onboard memory, a state machine... what else does it need to meet the classification of a CPU?

Originally Posted by SuperCyborg

Guess where the memory controller is on every modern computer ? Right, on the CPU!

Which means...? Intel CPUs already have multiple CPUs and vector CPUs on die. You act like adding another CPU onto the die is some amazing feat of linguistics.

Originally Posted by SuperCyborg

On a logicboard specifically designed to demo it and it was not a PCIe card.
Ever heard of cost management?
Its cheaper to R&D on a removable card than a logicboard. That way, only the card has to be remade vs the entire logicboard.

I'm sure it was for R&D.

That doesn't change that according to Intel's own diagrams for the final spec there isn't anything keeping Thunderbolt directly tied to the motherboard.

Originally Posted by SuperCyborg

TB uses the GPU built into the CPU.

Not on the iMac or Macbook Pro.

[SuperCyborg]

Originally Posted by goMac

It's not a general purpose CPU, but it's a CPU.

False. Please look up the definition of a CPU, most importantly what the letters C, P and U mean.

A North Bridge has a clock rate, onboard memory, a state machine... what else does it need to meet the classification of a CPU?

The ability to execute code given to it by software.

Not on the iMac or Macbook Pro.

False.
Both have an on-die GPU used by the TB multiplexer.
As mentioned before, the lack of on-die GPU is the only thing keeping TB from the MacPro.

[deconstruct60]

Originally Posted by SuperCyborg

False.
Both have an on-die GPU used by the TB multiplexer.
As mentioned before, the lack of on-die GPU is the only thing keeping TB from the MacPro.

Hugely flawed. On both iMacs and MBP 15", both the iGPU and the discrete GPU are hooked to the TB controller with a switch; The TB controller only sees two inputs from the switch. When the GPU workload is low the integrated GPU is used. On high GPU workloads the discrete GPU is used. The Thunderbolt port does not stop working when run at high GPU workloads. Therefore the discrete GPU is connected. Which one is a matter of the switch, it has nothing to do with Thunderbolt.

As long as there are DisplayPort signal that is all that matters.

The benefit that the iGPU provide is that there is no requirement for a discrete GPU for Thunderbolt to work in that context. That is extremely fundamentally different from saying the iGPU is required. The only requirement is that there is at least one DisplayPort output to use as input. That's it. Integrated in the CPU package or embedded on the motherboard both equally satisfy the "at least one" requirement.

It is highly questionable in a workstation whether want two GPUs hooked to the same "video input" on a Thunderbolt controller. Only one is necessary and the second can easily use conventional connections to a monitor. "looping back" from a removable card is not a requirement either.


P.S. there have been some TB concept boards that failed Thunderbolt certification but that seems more aligned with trying to use a removable PCI-e graphics card only to provide the video. Those are likely going to fail because a reconfigured system could just remove that card thereby removing a critical Thunderbolt input. That isn't going to pass. Again iGPU is convenient solution because inserting the CPU package means have one. If remove it the whole system isn't functional. A discrete solution would need to be equivalent in system crippling (e.g., desolder components off the motherboard) to likely qualify.
Intel is disallowing folks to "backdoor" creating a 'PCI-e data only' Thunderbolt variant. For PCs the socket has to provide both PCI-e data and DisplayPort video all the time. There would be no certified PCs where sometimes you plug in a monitor and it doesn't work. Peripherals are in a different boat with different expectations (especially when have just one Thunderbolt port).

[goMac]

Originally Posted by SuperCyborg

False. Please look up the definition of a CPU, most importantly what the letters C, P and U mean.


The ability to execute code given to it by software.

That's a general purpose CPU, not a CPU.

A CPU is not required to be programable. Although technically a northbridge can execute instructions given to it by the CPU, and is therefore programable, meeting your requirements...

Originally Posted by SuperCyborg

False.
Both have an on-die GPU used by the TB multiplexer.
As mentioned before, the lack of on-die GPU is the only thing keeping TB from the MacPro.

You're just flat out wrong here. The TB multiplexer on the iMac and Macbook Pro are connected to the discrete GPU and memory, not the CPU.

There is a Macrumors topic on someone trying to change this behavior, actually:
http://forums.macrumors.com/showthread.php?t=1140854

[SuperCyborg]

Originally Posted by deconstruct60

The TB controller only sees two inputs from the switch. When the GPU workload is low the integrated GPU is used. On high GPU workloads the discrete GPU is used.

Proof to back up your opinion?

----------

Originally Posted by goMac

That's a general purpose CPU, not a CPU.

False.

A CPU is not required to be programable.

False.
Those are called "microprocessors". Completely different from a CPU.

Although technically a northbridge can execute instructions given to it by the CPU, and is therefore programable, meeting your requirements.

False.

The TB multiplexer on the iMac and Macbook Pro are connected to the discrete GPU and memory, not the CPU.

False.

[goMac]

Originally Posted by SuperCyborg

False.

Needs citation.

[ScottishCaptain]

*eats popcorn*

I love this thread. One guy has a vocabulary of a single word, everyone else is busy trying to tell him how wrong he is. The next time I feel like trolling someone, I'm totally just going go reply "False." to every one of their statements and watch them reel.

-SC